Portfolio | Adhithya Laxman R G

Adhithya Laxman R G

I'm Adhithya Laxman R G

MSc. Data Science student at ETH Zurich, Switzerland. Passionate about Computer Vision, 3D Reconstruction, and Machine Learning — with hands-on experience in autonomous vehicle perception, edge AI deployment, and research-level deep learning.

Scroll Down

About me

I'm a MSc. Data Science student at ETH Zurich, currently doing research in 3D reconstruction and computer vision at the Computer Vision and Geometry Group. Before ETH, I completed my B.E. in Computer Science from Coimbatore Institute of Technology (top 2% of the class) and a Diploma in Data Science from IIT Madras — both in parallel. I genuinely enjoy working on hard ML problems, whether it's getting a model to run on edge hardware or digging into why a vision transformer fails on certain inputs.

Languages & ML Frameworks

Python C++ Java C SQL Shell Script PyTorch TensorFlow ONNX Scikit-learn NumPy Pandas Flask Django

Tools & Infrastructure

Docker Git Jenkins Azure DevOps Linux Kubernetes AWS S3 GCP Hive Spark ClickHouse MongoDB PostgreSQL Grafana

Focus Areas

Computer Vision 3D Reconstruction Deep Learning Edge AI NLP RAG Multimodal (Vision-Language) Transformers MLOps Distributed Systems Competitive Programming

ETH Zurich — MSc in Data Science

Zurich, Switzerland | Sep 2025 – Sep 2027

CGPA: 5.1/6

Relevant Coursework: Computer Vision, 3D Vision, Probabilistic AI, NLP, Deep Learning, Computational Statistics

Coimbatore Institute of Technology — B.E. Computer Science

Coimbatore, India | Nov 2021 – May 2025

CGPA: 9.57 — Top 2% of the class

IIT Madras — Diploma in Data Science & Applications

Chennai, India | Jan 2022 – Aug 2025

CGPA: 8.49 (Online Program, pursued in parallel with B.E.)

Research Student @ ETH Zurich – Computer Vision & Geometry Group

Zurich | Feb 2026 – Jul 2026

Improving robustness of a 3D reconstruction pipeline for in-the-wild indoor videos (HouseTour dataset, ICCV 2025) by integrating feed-forward reconstruction models to tackle a 40% failure rate in traditional SfM methods.

Benchmarking and adapting vision transformer models (DUSt3R, MASt3R, DAV3) to challenging indoor video data; built a hybrid SfM pipeline with COLMAP backend, with systematic evaluation workflows and multi-GPU ViT benchmarking.

Software Engineer Intern (CV / ML Systems) @ Caterpillar Inc.

Chennai | Jan 2025 – Jul 2025

Developed and deployed a monocular depth estimation pipeline for autonomous truck perception — trained on 50k+ images in PyTorch, optimized for edge AI hardware with camera-only sensor constraints.

Pushed model efficiency on edge hardware down to 4% depth perception error and cut inference time by 10ms, enabling reliable real-time inference in resource-constrained systems.

Applied MLOps practices (Git, Docker, Azure DevOps) to automate training and deployment pipelines end-to-end.

Software Engineer Intern (DevOps) @ Akamai Technologies

Bangalore | Jun 2024 – Aug 2024

Built automated ETL data pipelines using Jenkins to process large-scale data from Hive clusters handling over 1 billion records daily, integrated with ClickHouse and Grafana dashboards for real-time KPI visualization.

Migrated 6 legacy pipeline components (Python, ClickHouse, Jenkins jobs) to current versions, reducing pipeline failures and improving deployment reliability across the Release DevOps team.

Tech Stack: Python, ClickHouse DB, Grafana, Jenkins, ETL from Hive Cluster

Research Intern @ Samsung R&D Institute (PRISM)

Aug 2023 – Mar 2024

Led a team of 5 building Conditional GAN models in TensorFlow to generate MNIST-based video sequences. Hit 0.97 SSIM accuracy with a 15% improvement over baseline using custom pix2pix loss functions.

Awarded an Excellence Certificate for performance and the complexity of the implementation.

Tech Stack: Python, TensorFlow, OpenCV, Image Processing

Research Intern @ CSIR-CSIO

July 2023 – August 2023

Explored deep learning using Feed Forward Neural Networks for solar panel fault detection, achieving over 70% accuracy in classifying faults.

Ran comparative analysis across optimizers — Momentum, RMSprop, Adam, Nadam — to tune hyperparameters and squeeze out the best performance.

Investigating Relational Reasoning in Vision Language Models

PyTorch, Qwen 3-4B, Mechanistic Interpretability, CLEVR, ChartQA

Analyzing attention mechanisms in Qwen 3-4B VLM across 4 synthetic datasets using mechanistic interpretability (ablation, activation patching, circuit analysis) to identify relational reasoning limitations and knowledge shortcuts.

Evaluating cross-modal attention patterns with IoU and mutual information metrics to quantify visual-text alignment; applying Linear Probing across layers and heads to track shortcut learning in spatial reasoning.

Federated IDS with Deep Belief Networks [Springer 2025]

PyTorch, Python, Go, gRPC, FedAvg, Deep Belief Networks (RBM-pretrained), FNN

Led a team of 5 building a federated learning system for WiFi intrusion detection — privacy-preserving distributed training on 300k+ data points using a custom Go/Python gRPC architecture. Achieved 98% accuracy across heterogeneous networks.

Paper accepted at Springer 2025 (FEDDBN-IDS): Federated Deep Belief Network for Intrusion Detection on Heterogeneous WiFi Networks for previously unknown attacks.

Doc-Chat

Python, Flask, RAG, Google Generative AI Embeddings, ChromaDB, SQLite

Built an LLM-powered chatbot for PDF documents using a full RAG pipeline — PDF ingestion, semantic chunking, ChromaDB vector indexing, and grounded response generation. Supports Q&A, summarization, and quiz generation.

Blood Connect

Created a blood donation management system with a PostgreSQL backend and HTML/CSS frontend.

Lyric Lullaby

Built a music streaming app with authentication, playlist creation, and robust search functionality using Flask and SQLAlchemy.

Handwriting Letters – Classification

Developed a Feed Forward neural network for handwritten character recognition using the EMNIST dataset.

Achieved over 90% accuracy, demonstrating practical application in real-time character recognition.

Sentimental Analysis in R

Enhanced sentiment analysis accuracy through preprocessing, TF-IDF, and Naive Bayes classifier.

Achieved over 70% accuracy, showcasing applications for social media analytics.

Income Prediction – Adult Dataset

Implemented various machine learning techniques in Python for income prediction using the Adult Dataset from UCI.

Forecasted income based on demographic and employment-related features.

Task Management System

Developed a web-based application using PHP, Apache server, MySQL, HTML, CSS, and JavaScript for managing tasks.

Features include adding, editing, deleting tasks, and viewing detailed task information.

Library Management System

Python-based system using CSV files and a command-line interface for efficient library management.

Features secure user login, book management, and circulation records with a built-in fine calculation.

Springer 2025 — Accepted Publication

FEDDBN-IDS: Federated Deep Belief Network for Intrusion Detection on Heterogeneous WiFi Networks — accepted at Springer 2025.

Hackathon Wins

Winner — Internal Smart India Hackathon 2023 (1st out of 16 teams).

2nd place — Innovision24 Hackathon (2nd out of 48 teams).

A. John Wilson Endowment Prize

Awarded for the highest mark in Mathematics across all departments at Coimbatore Institute of Technology.

LeetCode Knight & Open Source

LeetCode Knight — Top 2.5% globally (Rating 2020+).

Open-source contributor to TheAlgorithms/Python (180k+ stars) — added neural network algorithms and optimization techniques.

Find Me

Email: adhithyarg26@gmail.com

Tel: +41 79 654 4680

Sonneggstrasse 17, Zurich, Switzerland

About Me